Research On Automatic Analysis Of Fabric Weave Patterns Based On Digital Image Processing

Improving efficiency and reducing cost is the key to enterprise survival in textile industry, and automatic detection of fabric weave patterns is the necessary foundation in fabric design and high-precision weave production, but traditional vision inspection is time-consuming, low efficiency, unable to meet the requirements of industry with low cost and high efficiency. Therefore, digital image processing and computer vision technology for classification and recognition of fabric weave patterns can realize the automatic and intelligent production of textile industry. This paper takes three basic woven fabrics as experimental objects, which include plain, twill and satin weave, and studies the two aspects of classification and recognition of fabric weave patterns using computer vision technology.On one hand, classify the fabric weave patterns. Firstly, fabric image is preprocessed by median filter and histogram equalization algorithm in order to filter noise and improve contrast. Then, the two approaches which are gray level co-occurrence matrix(GLCM) and Gabor filter are applied to extract the global and local texture features of fabric image. For the high-dimensional feature vector, principal component analysis algorithm is used to remove redundant information of feature vector to achieve dimension reduction. At last, an appropriate classifier of probabilistic neural network is used to training and testing the low-dimensional feature vector in order to achieve the automatic classification of three basic woven fabrics(plain, twill and satin weave). Experimental results indicate that the algorithm can classify fabric weave patterns accurately and efficiently, and obtain the best classification result(95%).On the other hand, recognize the fabric weave patterns. Firstly, fabric image is preprocessed by filter, histogram equalization and top-hat and bottom-hat transform algorithm in order to filter image noise and enhance the contrast of fabric yarn. Then, gray projection method is used to locate the yarn and segmented and obtained floats of fabric image. Meanwhile, unsupervised kernel fuzzy c-means clustering is used to recognize the floats based on the extraction of GLCM texture features of floats. The attribute of floats identified the according to the warp floats with relative high average gray mean value. At last, the improved distance matching function is employed to obtain the weave repeat which is used to correct error floats of preliminary identification results, and obtained more accurate identification result. Experimental results indicate that the algorithm can recognize the fabric weave patterns accurately and output accurate weave diagram.This paper realizes the efficient and automatic classification and recognition of fabric weave patterns, which have great theoretical significance and application value for production and detection the textile products.